Syslinux - Jul 2014 - [PATCH 00/05] Mending for mkdiskimage

On 07/10/2014 12:18 PM, Geert Stappers wrote:
>>
>> following is a series of patches derived from my proposed diff.
>> Based on syslinux-6.03-pre1, but should also apply down to 4.xx.
>>
>> I hope to have imitated the changelog style sufficiently.
>> Mine are a bit longer than usual with patches in the archive, i fear.
> 
> Yes, they are long.
> 
There seems to be no patch descriptions at all, which would really help
long-term maintainability.

	-hpa

Hi,

me:
> >> I hope to have imitated the changelog style sufficiently.
> >> Mine are a bit longer than usual with patches in the archive, i
fear.
Geert Stappers:
> > Yes, they are long.
H. Peter Anvin:
> There seems to be no patch descriptions at all, which would really help
> long-term maintainability.
It is a mid-term while ago since i proposed these patches.
Let me see how much i still understand my commit texts.

------------------------------------------------------------------------
1/5 seems clear enough. I'd propose to keep it:

  Option -s caused freshly created image files to be too small
  by a factor of 512.

------------------------------------------------------------------------
2/5 could need a longer story:

  Option -s caused a try to truncate the image file to the desired
  size. This failed with Linux block devices and caused lengthy
  zeroization of image size on such devices.
  This change ignores the failure of truncate() if option -s is
  given.

------------------------------------------------------------------------
3/5: 

  The size of a non-regular file was determined only for Linux
  block devices. All others caused an error exit.
  This change replaces the partly linux-specific file size
  determination by a sysseek() experiment. Tested on Linux, FreeBSD,
  and Solaris.

-------------------- 
Technical side note:
It turned out that the lseek(2) function of NetBSD does not work
with SEEK_END as expected from the manual. libburn had to resort
to a binary search with try-and-error to determine device sizes.
I did not check perl's sysseek() whether it suffers from the
lseek(2) shortcomming, but i would expect so.
It might be desirable to have an option for explicitely setting
a device size.

------------------------------------------------------------------------
4/5 seems clear:

  New option -S to sync zeroizing after each cylinder and to issue
  pacifier messages.

------------------------------------------------------------------------
5/5 is indeed quite obscure. How about this:

  Unclaimed space may remain after the end of the partition if the
  partition size stems from option -M, or is computed from arguments
  h and s because argument c is 0.
  Especially on device files, such space might contain a backup GPT
  which would be invalid after mkdiskimage but could lure partition
  editors into restoring the outdated main GPT.
  This change overwrites the space between partition end and image
  file end by zeros, unless option -s is given. (-s shall prevent
  zeroization in general.)

------------------------------------------------------------------------

If these descriptions are acceptable, do i have to re-submit the
patches, or can they be merged in by the committer ?
(I have the patches stored somewhere but would have to dig for them.)


Have a nice day :)

Thomas

On 07/11/2014 12:33 AM, Thomas Schmitt wrote:
> Technical side note:
> It turned out that the lseek(2) function of NetBSD does not work
> with SEEK_END as expected from the manual. libburn had to resort
> to a binary search with try-and-error to determine device sizes.
> I did not check perl's sysseek() whether it suffers from the
> lseek(2) shortcomming, but i would expect so.
> It might be desirable to have an option for explicitely setting
> a device size.
Perl sysseek() is just a wrapper around lseek(2).

However, there is such an option already... if you specify cylinder
count or, with the -M option, megabyte count.
> 5/5 is indeed quite obscure. How about this:
> 
>   Unclaimed space may remain after the end of the partition if the
>   partition size stems from option -M, or is computed from arguments
>   h and s because argument c is 0.
>   Especially on device files, such space might contain a backup GPT
>   which would be invalid after mkdiskimage but could lure partition
>   editors into restoring the outdated main GPT.
>   This change overwrites the space between partition end and image
>   file end by zeros, unless option -s is given. (-s shall prevent
>   zeroization in general.)
I have concerns about this, especially in the light of the above.  I
wouldn't want it to be the default at least.

	-hpa